【作者】 李雯；

【导师】 韩肖清；

【作者基本信息】 太原理工大学 ， 电气工程（专业学位）， 2020， 硕士

【摘要】 微电网作为一种有效消纳新能源的形式,在整个电力系统中具有良好的发展前景。在以光伏发电系统作为电源的微电网中,发电功率的准确预测对微电网优化调度和经济运行具有重要意义,是微电网能量管理系统的重要组成部分。本文基于微电网运行环境及监测数据实现对微电网中不同时间尺度下光伏发电功率的预测,其中日前整点发电功率预测采用DPK-means聚类和极限学习机(Extreme Learning Machine,ELM)相结合的方式实现,日内超短期滚动预测则是基于监测数据采用K-means聚类和随机森林(Random Forest,RF)的组合方法实现。综合不同时间尺度下的预测结果为微电网经济运行、方式调整和实时调度提供数据支撑。首先,在光伏发电模型基础上分析各环境因素对光伏发电功率的影响,进而确定预测所采用气象要素及历史数据。在对常用预测方法进行归纳总结之后确定采用聚类和智能算法相结合的预测思路。然后,针对日前预测寻找相似日时特征向量维度较高的问题,采用密度峰值法(Density Peaks Clustering,DPC)优化的K-means聚类方法,通过对初始聚类中心选择的优化提高聚类可靠性及稳定性,在此基础上采用相关系数作为聚类中相似程度的衡量标准,筛选出变化趋势相近的样本集。并通过需设置参数较少、泛化性能较强的极限学习机实现对次日整点光伏发电功率的预测。其次,在日内超短期预测部分,本文基于当前时刻气象数据和功率数据的监测情况,采用K-means和随机森林相结合的方式实现对光伏发电功率序列的预测。随着监测数据的更新实现预测数据的滚动更新,即时为微电网的优化调度提供预测结果。最后,实现微电网能量管理系统中发电功率预测模块的应用。将不同时间尺度的预测方法应用到光伏发电功率预测模块中,并实现该模块与其他各功能模块有机结合为微电网经济运行、方式调整和优化调度提供数据支撑。更多还原

【Abstract】 As a form of effective consumption of new energy,microgrid has good development prospects in the entire power system.For microgrid using photovoltaic as power sources,accurate prediction of photovoltaic power generation is an important part of the microgrid energy management system,which is of great significance for system scheduling and economic operation.Based on the actual environment and monitoring data of the microgrid.In this paper,the prediction of photovoltaic power generation in microgrid is realized at different time scales.The hybrid algorithm of DPK-means and Extreme Learning Machine(ELM)is used to achieve the next hour of full-time power generation in the day-ahead forecast.For intra-day power prediction,the combination of K-means clustering and Random Forest(RF)is used to achieve ultra-short-term rolling prediction of microgrid.The forecast results at different time scales provide forecast result to support the microgrid to adjust the operation mode and economic optimal dispatch.First of all,based on the photovoltaic power generation model,the influence of various factors on the photovoltaic power generation is analyzed,and then the meteorological elements and historical data used in the forecast are determined.After inductive summary of previous prediction methods,the prediction ideas using clustering and intelligent algorithms are determined.Secondly,it is necessary to find similar days for day-ahead prediction.In order to solve the problem of feature vectors with high dimensions,the K-means clustering method optimized by Density Peaks Clustering(DPC)is used to optimize the selection of the initial clustering center and improve clustering results.And the clustering results have better reliability and stability.On this basis,the correlation coefficient is used as a measure of similarity in clustering,and the sample set with similar trends is selected.By setting an extreme learning machine which has few parameters and strong generalization performance,the hybrid algorithm can get better generation power of the next hour.Then,in the intra-day ultra-short-term forecasting section,based on the monitoring of meteorological data and power data at the current moment,the combination of K-means and random forest is used to realize the power sequence prediction in this paper.With the update of the monitoring data,the rolling update of the forecast data is achieved,and the forecast results are provided for the optimal dispatch of the microgrid in a timely manner.Finally,the prediction methods at different time scales are used in the microgrid energy management system,and the functional modules of the microgrid energy management system are combined organically to ensure the reliability and economical operation of the microgrid.更多还原